
Manager
Manager connects the systems a restaurant already uses and turns scattered operations into a clearer picture of what needs attention next.
Context · decisions · executionSee the product →Systems
A Mycellium system begins with the operating reality: ownership, evidence, constraints, failure modes, and the decisions people still need to make.

Where Mycellium enters
A company can have capable teams and modern software and still run through broken handoffs, duplicated data, conflicting priorities, and decisions made without a shared view. The result appears as delay, rework, missed signals, and risk.
Mycellium enters at that boundary. We observe the real operation, locate where context and authority break down, and build the smallest system that can reconnect the work without hiding human judgment.
Current systems
Each architecture is selected for the field it has to survive.

Manager connects the systems a restaurant already uses and turns scattered operations into a clearer picture of what needs attention next.
Context · decisions · executionSee the product →
OKI Legal turns case documents into cited answers, review tables, workflows, and auditable decisions while keeping human review in control.
Documents · divergence · evidence · reviewSee the product →
A phone designed around your voice, so you can stay present. Make a call, send a text or check your day. Reach for the screen when you want it.
Voice · tasks · explicit authoritySee the product →
Sotatsu gives AI-native artists a place to publish work, build a presence and reach an audience without being reduced to content inventory.
Artists · curation · provenanceSee the product →
Hyperion brings rainfall, tide, terrain and urban exposure into one operational view, helping people see a fast-changing flood event earlier and more clearly.
Signals · risk · responseSee the product →
Korvo makes research, ideas and critical discussion about AI more accessible to Brazilian Portuguese readers · openly and without noise.
Research · learning · public recordSee the product →Open source
Small, public instruments from the lab. Read the code, then run them on your own work.

Cuts the tokens Claude Code, Codex and Grok Build spend, with your model and tools unchanged. On one real task per model (Oct 6, 2026), it used at least 42% fewer tokens on 6 of 9 OpenAI and 4 of 4 Grok models.
MIT · npm 0.5.6 · macOS and LinuxSee the product →Map the real workflow and its evidence.
Make decisions and human boundaries explicit.
Use only the architecture the situation needs.
What fails in the field becomes the next question.